Phones, Teenagers and the Evidence

Answer to the question

Is the link between smartphones and teen mental health real?

Pooled across 355,358 adolescents, digital technology use explains at most 0.4 per cent of the variation in well-being. Separate social media from total screen time and analyse girls alone, and the same three datasets give median coefficients from about minus 0.11 to minus 0.24. The population-level causal claim is neither established nor refuted.

Key takeaways

• Pooled across 355,358 adolescents, digital technology use accounts for at most 0.4 per cent of the variation in well-being, and that is the figure under most argument about smartphones and teen mental health.
• Separate social media from total screen time and analyse girls alone, and the same three datasets give median coefficients from about minus 0.11 to minus 0.24.
• No design capable of licensing a causal claim was run on nine- to fifteen-year-olds with smartphones: the trial is adults, the rollout study undergraduates in 2004 to 2006.
• Both camps accept that adolescent mental health has deteriorated and both want the platforms reformed, which is the part most coverage omits.
• School restriction, evaluated with mental wellbeing as a pre-specified primary outcome, made no difference, and the study cited most often in its favour measures exam results.

- Pooled across 355,358 adolescents, digital technology use accounts for at most 0.4 per cent of the variation in well-being, the figure under most argument about smartphones and teen mental health.
- Separate social media from screen time and analyse girls alone, and the same data give 1 to 6 per cent of the variation.
- No design that could license a causal claim was run on a child with a smartphone.
- Both camps accept that adolescent mental health has deteriorated, and both want the platforms reformed.
- The one evaluation with wellbeing as its pre-specified primary outcome, cross-sectional, found no difference.

At most 0.4 per cent of the variation in adolescent well-being goes with digital technology use, pooled across 355,358 adolescents, and that figure sits under most of what is said about smartphones and teen mental health. It is not the only defensible figure from the same data.

This piece carries a medical review line:.

What happened when schools restricted phones

The policy question on smartphones and teen mental health has been tested more directly than the causal one. Victoria Goodyear and thirteen colleagues compared 1,227 pupils aged 12 to 15 across 30 English secondary schools, 20 restrictive and 10 permissive. Mental wellbeing on the Warwick-Edinburgh scale was the pre-specified primary outcome. The adjusted difference was under half a point on a scale averaging 47, with an interval running from roughly two points below zero to one point above.

School-day phone time was lower in the restrictive schools, by about forty minutes, and in-school social media time too. Weekday and weekend totals showed no difference.

The design is cross-sectional rather than randomised, and the journal's own linked commentary records that this limits conclusions about causal effects and that only 10 per cent of invited schools took part. A null in an observational study does not demonstrate that restriction does nothing, only that the one evaluation with wellbeing as its pre-specified primary outcome found no difference.

Sara Abrahamsson's event study of Norwegian middle schools, peer-reviewed and ahead of print, reports no average effect of bans on education or mental health, with gender differences underneath: lower bullying for both sexes and, for girls, better grades and less take-up of psychological healthcare, a measure of service use rather than symptoms.

The largest evaluation is newer and not yet peer-reviewed. Hunt Allcott and five colleagues studied lockable pouches across the United States, using the largest provider's sales records. Pouches cut phone use. In the first year, disciplinary incidents rose and the authors report suggestive evidence that subjective well-being fell; the wellbeing estimates, in their words, appear to recover and become positive in later years.

The work quoted most often in favour of restriction measures something else: Louis-Philippe Beland and Richard Murphy, using the timing of schools' own bans in four English cities, found exam performance rising afterwards and no mental-health outcome in it.

Claims one and two: trend and timing

Four separable claims travel under one subtitle in Jonathan Haidt's The Anxious Generation. The first is that adolescent mental health deteriorated sharply from the early 2010s. Candice Odgers, whose Nature review, read here as an excerpted course copy, is the most prominent published criticism, writes that "we have a generation in crisis". Nothing we read disputes the trend. Attention and comprehension are a separate question, taken up in [Both Sides of the Screen-Time Argument](https://thewhyweproject.com/articles/screen-time-both-sides).

What the trend is measured on matters, and the three kinds of measure cannot be pooled in any account of smartphones and teen mental health. Symptoms reported on a questionnaire, diagnoses received and hospital admissions are three different quantities, sensitive to question wording, screening practice, help-seeking and recording rules. Haidt concedes part of this, writing that he is not saying none of the increase is due to a greater willingness to report these conditions or to adolescents pathologising normal anxiety.

The second claim is about timing: that the deterioration appeared in several countries at once, on a schedule matching the arrival of the smartphone and [a feed that pays for attention](https://thewhyweproject.com/articles/cruel-optimism-and-your-attention). Haidt puts it as an elimination: the only plausible theory he has found for the international decline is the change in teenagers' technology. That is an inference to the best explanation, and its force depends on the rivals having been ruled out. His own hedge sits on the same page: data is less abundant in non-Western nations and the patterns less clear.

Claim three: the size of the correlation

The third claim is that individual use is associated with worse mental health, and the numbers on smartphones and teen mental health are fought over here. Amy Orben, first author with Andrew Przybylski, ran a specification curve across three large datasets, two American surveys and the British Millennium Cohort Study, 355,358 adolescents. The method runs every defensible version of the analysis and reports the whole distribution. The association came out negative but small, at most 0.4 per cent of the variation in well-being, and in their judgement too small to warrant policy change.

Jean Twenge, with Haidt, Jimmy Lozano and Kevin Cummins, re-ran the method on the same datasets under four altered choices: social media separated from lumped screen time, the sexes separately, mediators dropped from the controls, scales weighted equally. Under the original settings they reproduced the original result. Under the revised ones, among girls, the median coefficients ran from about minus 0.11 to minus 0.24.

Those two quantities are not commensurable as stated. A coefficient of 0.11 is a little over 1 per cent of the variation and one of 0.24 about 6 per cent, against at most 0.4 per cent pooled, so the top of the revised range is about fourteen times the pooled figure rather than several times it. In coefficient terms, 0.4 per cent of variance is near six hundredths.

Both numbers are honest and both come from the same data.

The size of this association depends on how the data are cut, by more than the association itself.

One correction here is ours. The book's endnotes report that respecification as a single correlation coefficient. The published paper contains no such figure; it reports median coefficients across specifications, for girls, in the range above. That range is quoted from the paper's own abstract, read at source; its specification tables were not reachable on this pass.

What the pair establishes is that the size of this association depends on how the data are cut, by more than the association itself. The argument then ran in Nature Human Behaviour as a comment and reply, Twenge first on one and Orben first on the other, both paywalled and unread here.

Claim four: what the experiments measured

The fourth claim, and the only one the subtitle makes, is that these technologies caused the deterioration at population level. A correlation of any size cannot settle it; experiments and quasi-experiments can, and the question is what they used.

The best is a randomised trial by Hunt Allcott, Luca Braghieri, Sarah Eichmeyer and Matthew Gentzkow, which paid people to deactivate Facebook for four weeks before the 2018 American midterm election and found that deactivation raised subjective well-being. The participants were adults and the outcome reported well-being rather than a diagnosis.

The strongest quasi-experiment is Luca Braghieri's, with Ro'ee Levy and Alexey Makarin, which used the staggered arrival of Facebook across American colleges and found a negative effect on student mental health, through unfavourable social comparison. The exposure is Facebook between 2004 and 2006, on undergraduates, before the smartphone and Instagram. Haidt's notes send readers to a published critique of that study, unretrieved here, and explain why he still accepts the design.

Then the broadband work. Marta Golin instrumented German broadband access with features of the telephone network and reports results that are, in her word, suggestive: worse mental health for women aged 17 to 30, and not for men. The exposure is infrastructure rather than a phone. Haidt's handling of this class of study is a count of five pointing one way, none named in his main text and no magnitude given.

None of those designs was run on a child with a smartphone.

Where the two camps actually differ

Haidt's answer to the small-effect objection is that the critics' individual-level designs compare heavier and lighter users inside a population that has all been changed, and will therefore understate a collective transformation. In his endnotes he engages Orben and Przybylski with their own figures, granting that their headline number would be negligible if it were like-for-like and arguing that it is not. He designs the trial he says would settle it:

> "I know of no school district that has tested the hypothesis experimentally by randomly assigning middle schools to go phone-free while others make no change."

Odgers' case is that the rewiring claim is not supported by science, and that hundreds of researchers including herself searched for large effects and found a mix of no, small and mixed associations. Her sharpest point concerns direction: where associations over time are found, she reads them as showing that young people who already have mental-health problems use the platforms more, or differently. She calls many of the book's proposals reasonable, including stricter content moderation and designing for age, while judging age-based restrictions and device bans unlikely to work.

Her direction reading earns the same audit as Haidt's claims. For prior problems to explain the pattern, distress would have to come first and raise use, and the population trend be driven by something other than the technology. The evidence is thin in both directions: the longitudinal studies are the same observational panels whose sensitivity to analytic choices is described above, the trend sits on measures that changed over the same years, and the designs that fix the order of events used adults and undergraduates. Her reading is defensible and unestablished.

The gap is narrower than the coverage. Both accept the trend and both want the platforms changed, which is a question about [what technology does to who we are](https://thewhyweproject.com/articles/what-technology-does-to-who-we-are); they part on the direction of the longitudinal associations, on mechanism and on two named reforms. Odgers appears nowhere in the book's 366 pages.

Which leaves two worlds, and the evidence does not pick between them.

In one, the phone-based childhood is the major cause. In the other, heavy use marks distress that arrived first, and removing the device leaves it where it was. The reforms on offer assume the first.

What a parent can be told today

A parent reading about smartphones and teen mental health can still be told what this licenses. It licenses limiting heavy social media use, more so for a girl, since the association is present in every cut of the data and larger in some. It does not license the belief that a phone caused one child's distress, or that removing it will lift that child's wellbeing: the one evaluation with wellbeing as its pre-specified primary outcome found no difference, and the causal designs were run on adults and undergraduates.

It also licenses looking at the distress first, since the direction of those associations is exactly what is contested. What would change the answer is the trial Haidt specifies, and until it is run a parent is choosing under uncertainty.

Common questions

Is the link between smartphones and teen mental health real?

Real and negative, with two defensible sizes rather than one. Pooled, at most 0.4 per cent of the variation in well-being. Separate social media from screen time and look at girls alone, and the same data give 1 to 6 per cent, roughly fourteen times at the top. The causal claim is neither established nor refuted.

Do school phone bans improve teenagers' mental wellbeing?

The one evaluation with mental wellbeing as its pre-specified primary outcome found no difference between restrictive and permissive schools, in a cross-sectional design where a tenth of invited schools took part. A Norwegian study found no average effect. The largest American study, not yet peer-reviewed, reports suggestive evidence that wellbeing fell in the first year, with estimates the authors say recover and become positive later.

What would settle the smartphones and teen mental health question?

A randomised trial in schools, which Haidt specifies himself and says has never been run: middle schools assigned at random to go phone-free while others change nothing, whole schools as the unit, and outcomes agreed before the experiment begins. Nothing smaller will settle it.

Sources

Orben, A., & Przybylski, A. K. (2019). 'The association between adolescent well-being and digital technology use.' Nature Human Behaviour, 3(2), 173-182. doi:10.1038/s41562-018-0506-1

Twenge, J. M., Haidt, J., Joiner, T. E., & Campbell, W. K. (2020). 'Underestimating digital media harm.' Nature Human Behaviour, 4, 346-348. doi:10.1038/s41562-020-0839-4

Orben, A., & Przybylski, A. K. (2020). 'Reply to: Underestimating digital media harm.' Nature Human Behaviour, 4, 349-351. doi:10.1038/s41562-020-0840-y

Twenge, J. M., Haidt, J., Lozano, J., & Cummins, K. M. (2022). 'Specification curve analysis shows that social media use is linked to poor mental health, especially among girls.' Acta Psychologica, 224, 103512. doi:10.1016/j.actpsy.2022.103512

Odgers, C. L. (2024). 'The great rewiring: is social media really behind an epidemic of teenage mental illness?' Nature, 628, 29-30. doi:10.1038/d41586-024-00902-2

Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. (2020). 'The Welfare Effects of Social Media.' American Economic Review, 110(3), 629-676. doi:10.1257/aer.20190658

Braghieri, L., Levy, R., & Makarin, A. (2022). 'Social Media and Mental Health.' American Economic Review, 112(11), 3660-3693. doi:10.1257/aer.20211218

Golin, M. (2022). 'The effect of broadband Internet on the gender gap in mental health: Evidence from Germany.' Health Economics, 31(Suppl. 2), 6-21. doi:10.1002/hec.4570

Goodyear, V. A., Randhawa, A., Adab, P., Al-Janabi, H., Fenton, S., Jones, K., Michail, M., Morrison, B., Patterson, P., Quinlan, J., Sitch, A., Twardochleb, R., Wade, M., & Pallan, M. (2025). 'School phone policies and their association with mental wellbeing, phone use, and social media use (SMART Schools): a cross-sectional observational study.' The Lancet Regional Health - Europe, 51, 101211. doi:10.1016/j.lanepe.2025.101211

Weiss, L., & Bonell, C. (2025). 'School smartphone policies and adolescent wellbeing.' The Lancet Regional Health - Europe, 51, 101237. doi:10.1016/j.lanepe.2025.101237

Abrahamsson, S. (2026). 'Smartphone Bans, Student Outcomes and Mental Health.' Journal of Human Resources. doi:10.3368/jhr.0224-13403R2

Allcott, H., Baron, E. J., Dee, T., Duckworth, A. L., Gentzkow, M., & Jacob, B. (2026). 'The Effects of School Phone Bans: National Evidence from Lockable Pouches.' NBER Working Paper 35132. doi:10.3386/w35132

Beland, L.-P., & Murphy, R. (2016). 'Ill Communication: Technology, distraction & student performance.' Labour Economics, 41, 61-76. doi:10.1016/j.labeco.2016.04.004

Related concepts

• Pre-Specified Primary Outcome
• Vote Counting
• Quasi-Experiment
• Inference to the Best Explanation
• Specification Curve Analysis

People

• Candice Odgers
• Amy Orben
• Jean Twenge